The statistical paradigm: probabilistic and multivariate analysis applied through computational simulation in the interaction between genotype x environment

• Ano de defesa: 2019
• Autor(a) principal: Sarti, Danilo Augusto
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: eng
• Instituição de defesa: Biblioteca Digitais de Teses e Dissertações da USP
• Palavras-chave em Português: Ajuste probabilístico; Análise multivariada; Computational simulation; GxE interaction; Interação genotipoambiental; Multivariate analysis; Probabilistic fitting; Raspagem de dados; Scrapping; R programming; Simulação computacional; Software R
• Link de acesso: http://www.teses.usp.br/teses/disponiveis/11/11134/tde-18092019-090800/

Statistical analysis is based on an elementary paradigm and the relationship between probabilistic inductive inference, generation and validation of models, and the use of such information in decisions within a specific domain of knowledge. Additionally, techniques can be used to design specific experiments, such as the multi-environmental trials MET, to study the interaction between genotypes and environments. The fitting of probability distributions to data from phenomena allows the knowledge of the behavior of random variables and the later usage of such models in computational simulation. This procedure was carried out in the adjustment of models for maize grains weight, obtained via multi environmental trials. Several methods of adjustment of distribution and mixtures of normal distributions by the EM algorithm were used. The data were obtained through the use of scrapping with software R. Adjusted models were used to simulate, through computational methods implemented in language R, data with behavior known in parametric terms, generating a table that simulates the interaction between genotype and environment factors. Such simulated data were used to verify and compare models based on multivariate analysis, namely AMMI, weighted AMMI and GGE for the study of genotype environment interaction GxE. The results demonstrated the great effectiveness of the models in capturing the properties of the simulated data, contextualizing them as informational tools in the development of new products.